Biomedical Engineering Reference
In-Depth Information
in both the electrical potential on the surface of the head, the
electroencephalogram (EEG) and the magnetic field a few cen-
timeters beyond, the magnetoencephalogram (MEG), which is
the subject of this narrative.
MEG was introduced some 40 years ago
(1)
but stayed for
much of the time within small academic groups mainly in physics
departments, as a novelty area between physics and biology.
Although what was needed for the technology to become clini-
cally relevant was correctly identified early on
(2)
, the field reacted
rather slowly because the requirements were beyond what aca-
demic departments or small companies could afford. Two other
reasons delayed the emergence of MEG into modern neuroimag-
ing. First, the signal of an MEG channel after the usual processing
and averaging looks just like an EEG signal, so many saw MEG
as an expensive EEG technology. Some still do so today. Second,
it was known for well over a century that the mathematical prob-
lem of recovering the generators from the MEG and/or EEG
signal has no unique solution
(3)
. The non-uniqueness of the
bioelectromagnetic inverse problem is an undeniable theoretical
fact, but a rather benign problem in practice. Evidence that real-
time information about brain function was available from MEG
at not only excellent temporal resolution but also at fine spatial
detail became available in the 90s from novel analysis of multi-
channel MEG data
(4, 5)
. Eventually, helmet-like systems allowed
the mapping of the instantaneous magnetic field all around the
head in an instance. The analysis of the resulting signals provided,
for the first time, a view of dynamics of brain function across the
entire brain
(4)
.
From the numerous reviews of MEG, the 1993 work from the
Helsinki group remains the most comprehensive and informative
(6)
. More recent reviews have emphasized how, despite the issues
regarding the inverse problem, putative sources can be estimated
(7)
and, increasingly more often, how beam forming techniques
can usefully scan the source space point by point
(8)
.Mostof
the techniques discussed in the literature use linear methods for
extracting estimates for the generators. Heuristic analysis
(9)
and
theory
(10)
suggest that a specific form of non-linearity is neces-
sary for the solutions to possess expected properties for localized
distributed sources. Dealing successfully with the computational
penalty that goes with non-linearity, leads to reliable tomographic
estimates of brain activity from instantaneous MEG signals. Mag-
netic Field Tomography (MFT) is the name given to the result-
ing method of extracting estimates of brain activity
(11)
.MFT
solutions can scrutinize brain function at multiple spatiotemporal
scales. In the spatial domain, the range covers details a few mil-
limeters across (distinguishing activity within individual cytoarchi-
tectonic areas) to mapping across almost the entire brain. In the
time domain, events can be analyzed at timescales from a fraction
